---
title: Does the Vertex AI global endpoint break data residency for agents?
url: https://opencontroller.lyzr.com/blog/vertex-ai-global-endpoint-and-data-residency
published: 2026-11-14
updated: 2026-11-14
author: Lyzr
---

# Does the Vertex AI global endpoint break data residency for agents?

**For processing, yes. Google Cloud states that global endpoints do not provide regional isolation or data residency guarantees, and that requests may be processed in any Google Cloud location. Data you store at rest stays in the location you chose, but if a residency rule covers where prompts are processed, your agents must call a regional or `us`/`eu` multi-region endpoint instead.**

The global endpoint is attractive: higher availability, fewer capacity errors, TLS terminated close to the caller. Agent frameworks and code samples often default to it. That is how an agent built in one team can quietly undo a residency commitment made by another.

## What does Google actually guarantee for each endpoint type?

Google's data residency page separates two things: **data at rest** (for example tuned model weights and metadata) and **ML processing** (inference, tuning and training). Data at rest stays in the location you select. Processing depends on the endpoint you call:

| Endpoint type | Example | Where ML processing happens |
|---|---|---|
| Regional (locational) | `us-central1`, `europe-west4` | Within that region's multi-region or country jurisdiction |
| Multi-region (jurisdictional) | `us`, `eu` | Within the named jurisdiction |
| Global | `global` | Any Google Cloud location; no residency guarantee |

Three details matter for European teams:

1. The `eu` endpoint covers **EU member states only**. Google states that the UK and Switzerland are excluded.
2. For European regional endpoints, in-country processing **varies by model**. Some models process in-country (for example in France or Germany); others process within the EU boundary.
3. Not every model supports every endpoint. Google's page notes, for example, that `gemini-embedding-001` does not support the `us` or `eu` multi-region endpoints and is covered only through single-region endpoints.

## Why do agents end up on the global endpoint?

- **SDK defaults and samples.** Many quick-starts set `location="global"` to avoid capacity errors.
- **Model availability.** A team wants a model and finds it on the global endpoint, so they switch the whole agent.
- **Embeddings and tools.** The chat model is pinned to a region, but the embedding call in the retrieval tool is not.
- **Multiple projects.** Platform teams pin production, but experiment projects have no controls and later get promoted.

## How do you pin agents to a region?

Use this checklist for every agent that handles data with a residency requirement:

1. **Classify the agent.** Record the residency rule (for example "EU processing only") in the agent's registry entry.
2. **Pick the endpoint per model.** Check Google's deployments and endpoints page for each model the agent uses, including embedding models. If a model is not offered on a compliant endpoint, the agent cannot use it.
3. **Set the location explicitly** in code and configuration. Never rely on a default.
4. **Block global traffic with organisation policy.** Google's own guidance for IL5 customers is to restrict access to models on global endpoints through organisation policies. The same approach works for any residency rule. See [restricting Vertex AI models with org policy](/blog/restrict-vertex-ai-models-with-org-policy).
5. **Add a service perimeter.** VPC Service Controls stop data leaving the projects you define — see [VPC Service Controls for Vertex AI agents](/blog/vpc-service-controls-for-vertex-ai-agents).
6. **Prove it from logs.** Cloud Audit Logs record the endpoint and location of each call; sample them monthly. See [auditing Vertex AI calls made by agents](/blog/audit-vertex-ai-calls-made-by-agents).
7. **Re-test after model upgrades.** A new model version can change which endpoints are available.

## Does pinning the region make you compliant?

It gives you evidence, not a verdict. Google itself says multi-region endpoints are a foundation for regimes such as IL5 or ITAR, and that full compliance also needs customer-managed architecture, identity and access controls. Treat endpoint pinning as one control with logged proof, mapped to your obligation — for example GDPR transfer rules or a contractual residency clause.

## What about agents on other clouds?

The same pattern applies everywhere: residency for data at rest is usually easy; residency for processing depends on which endpoint or deployment type is called. Azure's global and Data Zone deployments and Amazon Bedrock's cross-region inference profiles have the same effect, each documented by the provider.

## How Opencontroller does this

Opencontroller auto-discovers agents, models and API keys across Google Cloud, AWS and Azure, and has a runtime adapter for Vertex AI. Its LLM Gateway puts control in the path, so you can route an agent's traffic only to approved regional endpoints and block anything else, with policies written as code. Posture flags risky configuration. See the [cross-platform governance guide](/blog/how-to-govern-ai-agents-across-platforms), the [control plane overview](/blog/what-is-an-ai-agent-control-plane), or [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Does data at rest leave my region if I use the global endpoint?**
Google states data at rest stays in the location you selected. The global endpoint affects where processing happens.

**Is a regional endpoint the same as in-country processing?**
Not always. A regional endpoint keeps processing within that region's jurisdiction; for some European regions and models that means the EU, not the single country.

**Should agents ever use the global endpoint?**
Yes, when no residency rule applies to their data. Record that decision in the agent's registry entry so it is visible at review.

## Key facts

- Google Cloud states that data stored at rest in a customer-selected location stays in that location, but ML processing location depends on the endpoint called.
- Regional endpoints keep ML processing within that region's multi-region or country jurisdiction (for example, us-central1 within the United States).
- The us and eu multi-region endpoints keep processing within that jurisdiction; Google says eu covers EU member states and excludes the UK and Switzerland.
- Google says global endpoints do not provide regional isolation or data residency guarantees; processing may occur in any Google Cloud location.
- gemini-embedding-001 does not support the us or eu multi-region endpoints, per Google's data residency page.
- Opencontroller has a runtime adapter for Vertex AI and auto-discovers agents, models and API keys across Google Cloud, AWS and Azure.
- Opencontroller starts from $250K per deployment per year.

## Related

- [How do you stop Vertex AI agents exfiltrating data with VPC Service Controls?](https://opencontroller.lyzr.com/blog/vpc-service-controls-for-vertex-ai-agents.md)
- [How do you restrict which models teams can use on Vertex AI?](https://opencontroller.lyzr.com/blog/restrict-vertex-ai-models-with-org-policy.md)
- [How do you audit Vertex AI and Gemini API calls made by agents?](https://opencontroller.lyzr.com/blog/audit-vertex-ai-calls-made-by-agents.md)
- [More on Platform Guides](https://opencontroller.lyzr.com/blog/cluster/platform-guides)
